Research On Key Technologies Of Mobile Application Traffic Classification And Identification Based On Conditional Generative Adversarial Network

With the arrival of a new era of mobile networks,the types and number of APPs on mobile devices are rapidly increasing,and the corresponding mobile traffic is also exploding.Traffic identification and classification is an important research content in the field of network management and network security.Operators need to perform dynamic access control based on the proportion of various types of traffic in the bandwidth to provide customers with a better user experience.The security supervision department needs to detect malicious traffic in the network traffic in real time to avoid serious losses.In the research of traffic classification,with the advent of port camouflage and random port technology,the traditional port-based traffic classification method quickly fails;the method based on payload characteristics also loses its effect because of the emergence of encrypted traffic.So some scholars began to apply machine learning to traffic classification problems,such as trying to use naive Bayes,SVM support vector machine,decision tree,random forest and other methods to extract feature and classification model training for flow statistical features or time series features.In these studies,feature extraction often relies on manual design by domain experts.The quality of feature design will have a greater impact on model learning.At the same time,as a manual method,it not only consumes time and effort,but also has limited ability to express the non-linear characteristics of flow under the condition of limited sample flow.Therefore,traditional machine learning methods rely heavily on artificial feature design and data sets.However,in the process of traffic collection,the captured traffic is often impure.It is necessary to filter out the required data in the complex network traffic and mark it.This process requires a lot of time and labor.At the same time,due to the different heat of various applications,it is often difficult for some unpopular applications to collect a large amount of traffic,which leads to data imbalance in the established data set.To this end,this paper designs a mobile traffic classification system based on unbalanced data sets.Firstly,a method for converting traffic into images is proposed.Without analyzing the data content,the data are converted into traffic images according to the grouping information.Then,a mobile traffic image sample generation model based on conditional generation of adversarial networks: Packet CGAN is proposed.Through the introduction of traffic image tags,the generation of is controlled to generate small-scale traffic image samples to balance the dataset.At the same time,the model structure of CGAN is adjusted,and CNN neural network is introduced as a traffic classifier.Finally,four traffic image classification models based on deep learning are used for traffic classification.In this paper,multiple sets of controlled experiments are designed to verify the feasibility of the Packet CGAN sample generation method.The experimental results show that the classification effect of the Packet CGAN balanced data set is better than other data sets.The main innovations of this article are as follows:1.A flow image conversion method based on flow grouping is designed.By analyzing the Pcap file,the flow is converted into a single-channel gray-scale flow map or a three-dimensional flow image without analyzing the specific content of the flow,so that the mature deep learning model in the field of image recognition can be applied to identify encrypted traffic.2.This paper designs a sample generate model of mobile traffic image generation based on Conditional Generative Adversarial Network: Packet CGAN.In the training process,the traffic image label is added as a condition variable to improve the "pattern collapse" phenomenon of traditional generational confrontation networks in a supervised learning manner.We adjust the network structure of the discriminator and generator.With the introduction of CNN neural network,while generating traffic samples,the Packet CGAN realizes the classification of mobile application traffic.